基于改进YOLOv5的轨道交通车厢乘客检测算法  

Passenger detection algorithm for railway carriages based on improved YOLOv5

作　　者：钱翱阳 沈辉[1] QIAN Aoyang;SHEN Hui(School of Mechanical Engineering,Yangzhou University,Yangzhou 225127,China)

机构地区：[1]扬州大学机械工程学院,江苏扬州225127

出　　处：《扬州大学学报(自然科学版)》2025年第2期38-44,62,共8页Journal of Yangzhou University:Natural Science Edition

基　　金：国家重点研发计划资助项目(2023YFD2300)。

摘　　要：针对通过人工检查和闸机信息获取乘客登车情况的效率偏低的问题,提出一种基于改进YOLOv5的轨道交通车厢乘客检测算法。模型将超标记注意力机制模块与YOLOv5算法的主干部分相结合,以提高算法对鱼眼图像特征的提取能力;在特征融合网络部分引入平行网络注意力机制模块,分析不同通道的特征重要度,以增强特征融合能力;将损失函数中的复杂交并比(complete intersection over union, CIoU)替换为更聚焦的最小点距交并比(intersection over union, IoU),以提高目标框的回归精度。结果表明,改进的YOLOv5算法在COF和LOAFs数据集上检测的平均精度较原始算法分别提高了0.037和0.016,能够有效完成轨道交通车厢鱼眼图像中的乘客检测任务。Aiming at the problem of low efficiency in obtaining passenger boarding status through manual inspection and brake information,a rail transit carriage passenger detection algorithm based on improved YOLOv5 is proposed.The model combines the superstandard memory attention mechanism module with the backbone of YOLOv5 algorithm to improve the extraction ability of the algorithm for fisheye image features.In the feature fusion network,the parallel network module is introduced to analyze the feature importance of different channels to improve the feature fusion capability.The complete intersection over union(CIoU)in the loss function is replaced with the more focused minimum point distance intersection over union(loU)to improve the regression accuracy of the target box.The results show that the average detection accuracy of the improved algorithm on the COF and LOAFs datasets is 0.037 and 0.016 higher than that of the original algorithm.It can effectively complete the passenger detection task in the fisheye image of rail transit carriages.

关 键 词：地铁车厢 乘客检测 YOLOv5 鱼眼图像 超标记注意力机制 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] U231.92[自动化与计算机技术—计算机科学与技术]